MBE Advance Access originally published online on March 31, 2006
Molecular Biology and Evolution 2006 23(6):1254-1268; doi:10.1093/molbev/msk015
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Research Article |
Transposon-Mediated Expansion and Diversification of a Family of ULP-like Genes
Department of Biology, McGill University, Montreal, Québec, Canada
E-mail: thomas.bureau{at}mcgill.ca.
 |
Abstract |
|---|
 TOP Abstract IntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Transposons comprise a major component of eukaryotic genomes, yet it remains controversial whether they are merely genetic parasites or instead significant contributors to organismal function and evolution. In plants, thousands of DNA transposons were recently shown to contain duplicated cellular gene fragments, a process termed transduplication. Although transduplication is a potentially rich source of novel coding sequences, virtually all appear to be pseudogenes in rice. Here we report the results of a genome-wide survey of transduplication in Mutator-like elements (MULEs) in Arabidopsis thaliana, which shows that the phenomenon is generally similar to rice transduplication, with one important exception: KAONASHI (KI). A family of more than 97 potentially functional genes and apparent pseudogenes, evidently derived at least 15 MYA from a cellular small ubiquitin-like modifier–specific protease gene, KI is predominantly located in potentially autonomous non–terminal inverted repeat MULEs and has evolved under purifying selection to maintain a conserved peptidase domain. Similar to the associated transposase gene but unlike cellular genes, KI is targeted by small RNAs and silenced in most tissues but has elevated expression in pollen. In an Arabidopsis double mutant deficient in histone and DNA methylation with elevated KI expression compared to wild type, at least one KI-MULE is mobile. The existence of KI demonstrates that transduplicated genes can retain protein-coding capacity and evolve novel functions. However, in this case, our evidence suggests that the function of KI may be selfish rather than cellular.
Key Words: genome evolution • gene duplication • transposable element • Mutator • Arabidopsis thaliana • SUMO
|
Introduction |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Transposons are abundant constituents of all eukaryotic genomes. Although they are considered "selfish DNA" because they survive not by phenotypic selection but through self-replication, mounting evidence indicates that transposons contribute by a variety of mechanisms to the function and evolution of host genomes (Makalowski 2003
; Kazazian 2004). For instance, eukaryotic transposons are able to duplicate and mobilize cellular gene sequences, potentially contributing to creative mutagenic processes like exon shuffling and gene duplication. Cellular genes flanking the 3' termini of retrotransposons can be duplicated by readthrough transcription initiated from the element, reverse transcription of mature mRNA, and insertion into genomic DNA (Moran, DeBerardinis, and Kazazian 1999; Elrouby and Bureau 2001). By a different but unknown mechanism, DNA transposons can incorporate unspliced fragments of unlinked cellular genes between the transposon termini in a process called transduplication (Juretic et al. 2005).
Mutator elements were first discovered in maize, and Mutator-like elements (MULEs) constitute a diverse superfamily of DNA transposons in plants, fungi, and prokaryotes (Le et al. 2000; Yu, Wright, and Bureau 2000; Turcotte, Srinivasan, and Bureau 2001; Lisch 2002; Chalvet et al. 2003; Neuveglise et al. 2005). Autonomous MULEs contain a mudrA gene which encodes a transposase required for mobility. In most organisms, MULEs have long, high-identity terminal inverted repeats (TIRs), but roughly one-third of Arabidopsis MULEs do not. It was not previously known whether non-TIR MULEs are capable of transposition (Le et al. 2000; Yu, Wright, and Bureau 2000).
Maize Mutator elements were first observed to contain insertions of non-Mutator DNA (Chandler, Rivin, and Walbot 1986), and transduplication has subsequently been documented in Arabidopsis MULEs (Le et al. 2000; Yu, Wright, and Bureau 2000), CACTA elements in Japanese morning glory (Kawasaki and Nitasaka 2004) and soybean (Zabala and Vodkin 2005), as well as maize rolling circle elements (Morgante et al. 2005). Two genome-wide studies recently identified over 1,300 duplicated gene fragments formed by transduplication (i.e., transduplicates) in rice MULEs (Jiang et al. 2004; Juretic et al. 2005), but despite their abundance all existing transduplicates in rice appear to be pseudogenes (Juretic et al. 2005), and there is, to our knowledge, no documented case of a transduplicated gene encoding a functional protein. It has also yet to be determined whether transduplicates have other functions, such as the provision of sequence reservoirs for gene conversion or the generation of small RNAs (sRNAs) which might participate in RNA-mediated silencing of paralogous cellular genes.
Here we report the results of a genome-wide survey of MULE-mediated transduplication in Arabidopsis and the discovery of one highly unusual case. We find that the Arabidopsis genome contains at least 97 sequences containing strong similarity to peptidase C48, a conserved domain (CD) found exclusively in ubiquitin-like protein–specific protease (ULP)–like genes. Most of these sequences are located in predicted genes, but our analysis indicates that only eight are cellular ULP-like (AtULP) genes and the remainder form a unique family of transduplicates located in non-TIR MULEs which we name KAONASHI (KI; named after the mysterious character in Hayao Miyazaki's animated film "Spirited Away" who is split between conflicting worlds and identities). We contrast the characteristics of KI with other transduplicates; examine KI phylogeny, conservation, and age; compare KI expression patterns to those of mudrA and cellular AtULP genes; and investigate KI-MULE mobility and transcriptional silencing. We conclude by arguing that KI is a functional and possibly selfish gene family and discuss its significance in understanding the evolutionary forces underlying the phenomenon of transduplication and transposon evolution in general.
|
Materials and Methods |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Sequences
The Institute for Genomic Research (TIGR) Arabidopsis thaliana (Columbia-0) (hereafter referred to simply as Arabidopsis) genome sequences and annotations version 5.0 (TIGR5) were accessed both locally and online at http://www.tigr.org (AGI 2000
; Haas et al. 2005). Pseudochromosome sequences were downloaded from http://www.tigr.org on June 2004.
Identification of MULEs, Transduplicates, and Peptidase C48
MULEs are characterized by long terminal sequences (greater than 100 bp) which are conserved within families, flanked by short direct repeats (9–11 bp) called target site duplications (TSDs) which are generated on insertion. MULE internal sequences are variable even between closely related elements due to deletions, insertions (including transduplications), and substitutions and also because only some MULEs contain a mudrA gene. We modified a previously described automated in silico procedure (Juretic et al. 2005) to identify MULEs in the TIGR5 Arabidopsis genome sequence, augmented with extensive manual curation, briefly described here. We compiled a library of terminal sequences (100 bp in length) from a set of previously characterized MULEs (Le et al. 2000; Yu, Wright, and Bureau 2000) and additional MULEs identified through manual curation. Using these as queries, we identified a complete set of putative MULE termini through similarity searches with the genome sequence using WU-Blast (version 2.0 release March 27, 2004; http://blast.wustl.edu) with MASKERAID (Bedell, Korf, and Gish 2000) as a search engine for REPEATMASKER (version 2004/03/06; options: nolow, nocut, no_is, s; http://www.repeatmasker.org). Pairs of termini from the same family, with correct orientation, separated by less than 30 kbp were matched, and the immediate flanking region was searched for 9-bp TSDs. Matched pairs of termini flanked by TSDs with at most two mismatched base pairs were considered to be intact MULEs.
Each MULE subsequently found to contain a transduplication or peptidase C48 sequence (see below) was verified by manual inspection of its termini, repetitiveness, and TSDs. Greater than two TSD mismatches were permitted, subject to manual inspection, if the MULE contained a peptidase C48 sequence or a transduplicate. MULEs were categorized as TIR or non-TIR according to the previously described library, in which TIR MULEs were defined as having at least 60% nucleotide identity in an alignment between the 5' terminal 100 bp and the reverse complement of the 3' terminal 100 bp (Yu, Wright, and Bureau 2000). We then identified additional MULEs surrounding mudrA sequences not contained in MULEs in this data set and iterating the above procedure. The distribution of MULEs on Arabidopsis chromosomes was visualized using the Nottingham Arabidopsis Stock Centre Arabidopsis Ensemble KaryoView tool (http://genome.arabidopsis.info).
To locate candidate transduplicates, we searched for cellular CDs within the MULEs by identifying all CDs and then filtering out CDs which are found in MULEs or other transposons. Because many transduplicates will probably not contain a CD, this method is expected to have a high false negative rate; however, it has the advantage of a low false positive rate (see Discussion). Putative transduplicated CDs were identified by querying the National Center for Biotechnology Information (NCBI) CD-Search (default settings; accessed July 2004; Marchler-Bauer and Bryant 2004) with six-frame conceptual translations of the MULE sequences. Transposon-related CDs were filtered out and ambiguous CDs, which may have been derived either from a transposon or a cellular gene, were also filtered out unless adjacent to a nonambiguous cellular CD. Excluding peptidase C48 (pfam02902) which was analyzed separately (see below), a putative cellular gene paralog for each candidate transduplicate was identified as the locus corresponding to the second best hit (best non–self-hit) in an NCBI BlastN (Altschul et al. 1990) search of the genome sequence.
A large number of MULEs were found to contain the peptidase C48 domain. To exhaustively locate all Arabidopsis sequences similar to peptidase C48, we employed two complementary methods: (1) CD-Search of TIGR5 annotated open reading frames (ORFs) and (2) PSI-TBlastN (version 2.2.8; Schaffer et al. 2001) search of the TIGR5 genome sequence, using as query a consensus of eight representative Arabidopsis peptidase C48 sequences computed by the NCBI Conserved Domain Database (gi|3377828, gi|5731755, gi|4309748, gi|3377837, gi|4678213, gi|3859612, gi|3080361, and gi|4733978). We ignored annotated exon and gene boundaries and extracted maximum length peptidase C48 sequences from genomic sequences at the positions identified in these searches and, because many sequences were apparent pseudogenes, we disregarded frameshifts and stop codons. Frameshifts were defined as two consecutive ORFs (putative exons) in different frames separated by a gap (putative intron) of less than 10 bp, a conservative threshold given that 99.9% of true Arabidopsis introns are longer than 50 bp (Yu et al. 2002). Frameshifts and premature stop codons in the peptidase C48 domain were counted. Cases where peptidase C48 sequences were not within a MULE were reexamined in an attempt to identify evidence of MULE-like sequences in the flanking genomic regions. Neighboring mudrA genes were identified based on TIGR5 annotations. The positions and distribution of peptidase C48 sequences were visualized using The Arabidopsis Information Resource (TAIR) Chromosome Map Tool (http://www.arabidopsis.org; Garcia-Hernandez et al. 2002). In four cases, where there was no TIGR5 locus at the position of the peptidase C48 sequence, we assigned ad hoc identifiers based on genomic positions (e.g., At4k0284 was used to identify a peptidase C48 sequence located at chromosome 4 in the region 28.4–28.5 Mbp).
Estimates of the number of peptidase C48 domains in various species were obtained from the Protein Families database (Pfam; version 19.0; Bateman et al. 2004). We also searched for KI-like sequences in preliminary Brassica oleracea genome sequence contigs (less than 0.5x coverage) using a TBlastN search on the TIGR Web site (http://www.tigr.org; accessed 10/2005) with a representative KI peptidase C48 domain (from At2g12100) as the query. We verified that the resulting B. oleracea sequences contained peptidase C48 using NCBI CD-Search.
Alignment, Phylogeny, and Conservation
3DCOFFEE (standalone T-COFFEE version 2.50; option: special-mode = 3dcoffee; O'Sullivan et al. 2004), a highly accurate multiple sequence alignment tool, was used to align all Arabidopsis peptidase C48 sequences with sequences from Drosophila melanogaster (Ulp1 gi|18860521:1318–1508), Saccharomyces cerevisiae (Ulp1 gi|6325237:433–617; Ulp2 gi|6322158:444–673), Schizosaccharomyces pombe (gi|2894265:377–564), Homo sapiens (Senp2 gi|54607091:397–586), and human adenovirus type 2 protein (HavULP; gi|34810217:50–144), as well as three-dimensional structure data from S. cerevisiae Ulp1 (Protein Database [PDB] ID 1EUV; Mossessova and Lima 2000) and H. sapiens Senp2 (PDB ID 1TH0; Reverter and Lima 2004). Alignments were manually edited to correct the alignment of the invariant glutamine in HavULP and KI sequences as in Mossessova and Lima (2000). We used the Phylogeny Inference Package (version 3.6; http://evolution.genetics.washington.edu/phylip.html) programs SEQBOOT, PROTDIST, NEIGHBOR, and CONSENSE to construct an extended majority rule consensus tree from 1,000 bootstrap replicates. We also clustered both protein and DNA sequences using NCBI BLASTCLUST (standalone version 2.2.10; http://www.ncbi.nlm.nih.gov/BLAST) to define clusters at various levels of divergence. Both the phylogenetic and clustering analyses supported a division between eight previously documented cellular AtULP genes (Novatchkova et al. 2004), which are not located in MULEs, and the remaining sequences, namely KI genes, which are located in MULEs.
To prepare the alignment for synonymous substitution rate analysis, we made the following adjustments. Non-Arabidopsis sequences were removed. Genomic DNA sequences were substituted for corresponding amino acid sequences, stop codons were replaced with gaps, codons containing gaps in more than 15% of sequences were removed, and clusters of greater than 88% amino acid identity were pruned to one sequence. Two alignments were created, one that included cellular AtULP genes and a second that excluded them. TREEVIEW (version 1.6.6; Page 1996) was used in making corresponding edits to the consensus tree. The adjusted alignments and trees were used to estimate dN/dS (the ratio of nonsynonymous to synonymous nucleotide substitutions per site) using the Phylogenetic Analysis by Maximum Likelihood package (PAML; version 3.14 release January 2004). BASEML was used to estimate initial branch lengths and CODEML was used for dN/dS calculations (default parameters except CodonFreq = 2, clock = 0, cleandata = 0, fix_blength = 1, and model, NSSites, and fix_omega as below). Overall dN/dS values for KI and AtULP peptidase C48 domains were calculated using the several-ratios branch model (model = 2, NSSites = 0; Yang 1998; Yang and Nielsen 1998) with one dN/dS value for KI and a second for AtULPs. KI clade-specific dN/dS values were calculated under the same model, using only KI sequences (no AtULP), two runs per clade (one run with fix_omega = 0, one with fix_omega = 1 and omega = 1), with one dN/dS value for the selected clade and a second for all remaining nodes. Likelihood ratio tests were used to determine whether clade-specific dN/dS values differed significantly from unity. In addition to the above calculations, dN/dS was calculated for various other sequence subsets and tree branches (e.g., excluding outliers, including nearly identical sequences, including or excluding AtULP genes) using various models and parameters with similar results (data not shown).
Expression
We used a combination of data sources to investigate the expression of KI, mudrA, and cellular AtULP genes. We identified expressed sequence tag (EST) and full-length cDNA sequences using the Munich Information Center for Protein Sequences Arabidopsis thaliana Database (http://mips.gsf.de/proj/thal/db; Schoof et al. 2002) and TAIR (http://www.arabidopsis.org; Garcia-Hernandez et al. 2002), and we also searched for evidence of expression in the whole-genome tiled oligonucleotide array data of Yamada et al. (2003). Using GENEVESTIGATOR (https://www.genevestigator.ethz.ch; Zimmermann et al. 2004), we collected microarray measurements (ATH1 22k array, Columbia-0) in different tissues from a pooled and standardized database incorporating several large public databases (Craigon et al. 2004; Barrett et al. 2005; Parkinson et al. 2005). Finally, 17-bp mRNA and sRNA massively parallel signature sequencing (MPSS) data were extracted from the Arabidopsis MPSS database (http://mpss.udel.edu; Nakano et al. 2006).
Mobility
Plant Material
Mutant seeds were obtained from Tetsuji Kakutani (Department of Integrated Genetics, National Institute of Genetics, Japan). Plants were raised in a growth chamber under 24 h daylight at 22°C. Genomic DNA was isolated from leaf tissue using the DNeasy Plant Mini Kit (QIAGEN, Mississauga, Ontario, Canada).
Transposon Display Analysis
Transposon display (TD) was performed as described by Wright et al. (2001) with minor modifications. Genomic DNA (100 ng) was digested with 2.5 U BfaI (New England Biolabs, Beverly, Mass.) and ligated to 15 pmol adaptor cassettes (5'-TAGCAAGGAGAGGACGCTGTCTGTCGAAGGTAAGGAACGGACGAGAGAAGGGAGA-3' and 5'-TCTTCCCTTCTCGAATCGTAACCGTTCGTACGAGAATCGCTGTCTCTCCTTGC-3') with T4 DNA ligase (Invitrogen, Carlsbad, Calif.). The ligation reaction was diluted fourfold. A 3-µl aliquot of diluted ligation mixture was used as template for preselective amplification with MULE-specific primer MuP-1 (5'-GGTCAGTTTTTGGCT(G/T)AATGGCTAA-3') and adaptor-specific primer ap-1 (5'-CGAATCGTAACCGTTCGTACGAGAATCGCT-3') and the following polymerase chain reaction (PCR) conditions: initial denaturing of 10 min at 94°C; 20 cycles of 1 min at 94°C, 1 min at 55°C, and 1 min at 72°C; and final extension of 10 min at 72°C. PCR products were diluted 100-fold in MilliQ distilled water. A 3-µl aliquot of diluted PCR products was used as template for selective amplification under the same reaction conditions with nested element-specific primer MuP-2 (5'-AAAGTGGGTCAA(C/T)GGCTACTG-3') and IRD700-labeled (LiCor, Lincoln, Nebr.) nested adaptor-specific primer ap-2 (5'-GTACGAGAATCGCTGTCCTC-3'). Five microliters of loading dye was added to the final amplification products, which were separated by size and visualized on a 5.5% denaturing polyacrylamide gel using a LiCor IR4200 DNA sequencer. LiCor IRD700-labeled DNA (50–700 bp) was used as molecular weight markers. Polymorphic DNA fragments were isolated from the gel using the same primers (MuP-2 and ap-2) cloned into pCR 2.1 vectors (TA cloning kit, Invitrogen, Canada) and sequenced using a 3730xl DNA Analyzer (Applied Biosystems, Foster City, Calif.) at the McGill University and Genome Québec Innovation Centre (Montreal, Canada).
Reverse Transcriptase–PCR
Total RNA was isolated from floral tissue using the RNeasy Plant Mini Kit (QIAGEN, Germany) and treated with RNase-free DNase (QIAGEN) to eliminate contaminating DNA. cDNA was synthesized from 1 µg of total RNA using the SuperScript III First-Strand Synthesis System for reverse transcriptase–PCR (RT-PCR) following the manufacturer's instructions. PCR was performed using the following conditions: one cycle of 2 min at 94°C; 35 cycles of 30 s at 94°C, 30 s at 58°C, and 90 s at 68°C; and a final extension of 5 min at 68°C. PCR products were separated on a 1.5% agarose gel containing 0.4 µg/ml ethidium bromide. Forward (F) and reverse (R) primer sequences used for the amplification of KI-At2g12100 and mudrA-At2g12150 transcripts were as follows: UPF, 5'-GAAGAGAAATCGGTATGTCGTT-3'; UPR, 5'-GACGTTGCAGGCATATAGCT-3'; UP1F, 5'-GCAACTGGTAGTCTTCCTGTC-3'; UP1R, 5'-ACAACTTCTTCTTCAGGATTT-3'; UP2F, 5'-GTTCAGAAAGGACTTGGTGGA-3'; UP2R, 5'-ATTAAGGCATTCTCCCTTGGA-3'; MPF, 5'-GAACATGAGGACGAGGATAAC-3'; MPR, 5'-CTGTTGTCACTAGACCGTCA-3'. The actin gene ACT4 was used as control with primers Act4-U (5'-GCATAGAGTGAGAGAACAGC-3') and Act4-D (5'-GACGTTGAAGACATTCAACC-3').
|
Results |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Detection of MULEs and Transduplicates
The MULE-mediated acquisition of cellular gene fragments in Arabidopsis was previously documented in a survey of approximately 15% of the genome (Yu, Wright, and Bureau 2000
). To better characterize MULE-mediated transduplication in Arabidopsis, we modified a procedure previously shown to be accurate (Juretic et al. 2005) and identified 722 MULEs (table ST1, Supplementary Material online). Consistent with earlier reports (AGI 2000; Le et al. 2000; Yu, Wright, and Bureau 2000; Singer, Yordan, and Martienssen 2001), MULEs were found to be distributed across all chromosomes and concentrated at higher densities in pericentromeric regions and heterochromatic knobs (table ST1 and figure SF1, Supplementary Material online). To locate transduplicate candidates, we searched for CDs within the MULEs, filtering out mudrA-related CDs. To differentiate between transduplication and transposon insertion, which occurs frequently, we also filtered out CDs characteristic of other transposons. Because most transduplicated sequences undergo frequent mutation, we identified CDs by scanning conceptual translations of MULE internal sequences in all six reading frames, ignoring frameshifts and stop codons. Each candidate transduplicate-containing MULE was manually verified, corresponding full-length cDNAs and ESTs were identified, and cellular genes corresponding to transduplicates were located and characterized (table 1; table ST2, Supplementary Material online). We found a total of 86 CDs in 22 transduplicated sequences and 19 MULEs. However, because our methods only detect transduplicates that contain a CD, these almost certainly constitute only a subset of Arabidopsis transduplicates. Alignments of transduplicated sequences to putative paralogous cellular genes had nucleotide identities ranging from 58% to 98% (unweighted mean 81%). Sixteen MULEs contained a single transduplicate and three MULEs contained two transduplicates each. Nineteen of the transduplicated CDs were present in two groups of MULEs, at the following locations: chr1|16662129–16666762, chr3|11796038–11800933, chr3|14747509–14753917, chr3|20780577–20799426 and chr4|591671–592987, chr4|1866831–1882462, chr4|5200471–5204006. Eighty-one percent of transduplicated CDs were truncated by at least 30%.
|
Identification of KI
One apparently transduplicated CD, peptidase C48, had markedly unique characteristics. While most transduplicates were single copy and highly truncated, peptidase C48 was largely intact and found in a large number of MULEs (table ST3, Supplementary Material online). Peptidase C48 is a cysteine protease domain approximately 200 amino acids in length found in Ulp-like proteins, located at the C-terminus of Ulp1 and in the central region of Ulp2 (Li and Hochstrasser 1999
, 2000). To fully characterize the occurrence of peptidase C48 in Arabidopsis, we scanned the genome sequence and located 84 peptidase C48 domains truncated by less than 5%, an additional 21 truncated by less than 20% (table ST3, Supplementary Material online), and at least an additional 35 truncated by at least 20% (table ST4, Supplementary Material online). We refer to domains truncated by 20% or less as "intact" and to the remainder as "truncated" and implicitly refer to intact domains unless otherwise noted. Eight of the intact domains were located in nonpseudogenic TIGR5 ORFs previously identified as probable AtULP genes (Novatchkova et al. 2004) and were not associated with MULE features such as adjacent mudrA ORFs or flanking MULE termini. We refer to these eight ORFs as cellular AtULP genes. Sixty-nine of the remaining 97 intact domains were located in intact MULEs. Note that we use the term "intact MULEs" to mean not truncated, that is, MULEs which have matching termini and TSDs (see Materials and Methods). This is different from "autonomous," which would require that, in addition to being intact, a MULE contains a functional mudrA gene and is furthermore capable of mobilizing itself in the absence of other MULEs. Nevertheless, most of these intact MULEs also contained a mudrA gene, many of which (29 of 69) were in turn found to contain all three mudrA-related CDs (pfam03108, pfam00872, and smart00575). These MULEs are potentially autonomous (table ST3, Supplementary Material online). The remaining 28 intact peptidase C48 domains appeared to be located in truncated MULEs as they were highly similar to sequences in intact MULEs and were usually associated with one or more MULE features such as a single unpaired terminus or a mudrA ORF.
Ninety-three of these 97 intact peptidase C48 domains were located in TIGR5 annotated ORFs (table ST3, Supplementary Material online). These MULE-related sequences constitute a novel family of ULP-like genes, which we named KI. Seventy-seven percent of KI-MULEs contained a mudrA gene and 60% contained one or more additional ORFs, which may be transduplications or transposon insertions (table ST3, Supplementary Material online). KI and mudrA are located on opposite strands in convergent orientation. Interestingly, all KI-MULEs have the unusual characteristic that their termini do not form high-identity inverted repeats, that is, they are non-TIR MULEs (Yu, Wright, and Bureau 2000).
Phylogeny and Age
To investigate the phylogeny of KI, we performed multiple sequence alignments of the predicted amino acid sequences of all intact Arabidopsis peptidase C48 domains including both KI and cellular AtULP genes, representative sequences from diverse species, and three-dimensional X-ray crystallographic structures of S. cerevisiae Ulp1 (Mossessova and Lima 2000) and H. sapiens Senp2 (Reverter and Lima 2004; fig. 1; figure SF2, Supplementary Material online). We constructed a neighbor-joining tree based on protein distances (fig. 2; figure SF3, Supplementary Material online). KI and the eight cellular AtULP genes formed two highly diverged phylogenetic groups. Although the bootstrap support for each major KI clade (see below) is 100%, it is below 50% for the most ancient nodes due to the high degree of divergence between clades and between KI and AtULP genes. This indicates that there is not enough evidence either to determine which cellular AtULP is most closely related to the KI family or to evaluate the order in which KI clades diverged from one another.
|
|
Similarity-based clustering of the amino acid sequences was consistent with the phylogenetic analysis and permitted further definition of KI subgroups. At an arbitrary threshold of 1.0 bits/residue (approximately equivalent to 45% identity), we grouped KI into nine clades of 3–20 members, leaving four single-member cluster outliers. At the same threshold, cellular AtULP genes formed four groups (At4g00690, At4g15880, At3g06910; At1g60220, At1g10570; At4g33620, At1g09730; and At5g60190), consistent with our phylogenetic tree and with previous studies (Kurepa et al. 2003
; Novatchkova et al. 2004). The non-Arabidopsis Ulp-like sequences each formed single-member clusters at this threshold. Further clustering at various thresholds identified additional subgroups. For instance, at 100% nucleotide identity there were 2 clusters of 2 and 4 members, at 99% nucleotide identity there were 8 clusters of 2–5 members, and at 88% amino acid identity there were 13 clusters of 2–15 members, distributed across all clades (figure SF3, Supplementary Material online). To estimate a minimum age of KI formation, we searched for sequences similar to a representative KI peptidase C48 domain (At2g12100, Clade 9) in preliminary B. oleracea genomic contigs (less than 0.5x coverage) and identified 228 KI-like sequences (E < 10–10). The B. oleracea sequences were most closely related to Arabidopsis Clades 7 and 8 with maximum 33% identity in a 208–amino acid BlastP alignment.
Conservation
Peptidase C48 contains a putative catalytic triad of histidine, aspartate, and cysteine as well as a highly conserved glutamine residue positioned near the active site (Li and Hochstrasser 1999, 2000). We examined the KI peptidase C48 domains to determine whether these features were conserved and also whether the domains contained obvious disablements such as frameshifts or premature termination codons. Although many of the KI peptidase C48 domains were found to have one or more obvious defects, as expected for transposon-related ORFs, 53 had no obvious disablement and were positioned at the C-terminus like in Ulp1. In three of the KI clades (Clades 2, 3, and 9), the majority of sequences encoded all four invariant residues; in two clades (Clades 7 and 8), the majority of sequences encoded all three residues in the catalytic triad (histidine, aspartate, cysteine) but not the putative invariant glutamine; in three clades (Clades 4, 5, and 6), the majority of sequences encoded histidine, aspartate, and glutamine but not cysteine; and in the remaining clade (Clade 1), the majority of sequences encoded the invariant glutamine but none of the catalytic triad. It is possible that some of the apparently missing invariant residues are actually present but were improperly aligned; however, the regions adjacent to the invariant sites are particularly well conserved making this unlikely (fig. 1; figure SF2, Supplementary Material online).
To evaluate whether the amino acid sequences of KI genes had been subject to selective constraint, we estimated peptidase C48 domain dN/dS ratios for the entire KI subtree, for each KI clade, and for cellular AtULP genes using maximum likelihood (fig. 2; table ST5, Supplementary Material online). The overall dN/dS for KI and AtULP genes were 0.24 and 0.12, respectively. The dN/dS values of eight of the nine clades ranged from 0.11 to 0.51 and were all significantly smaller than unity (likelihood ratio test, P < 0.001). The only exception was Clade 2, which had dN/dS of 1.35, a value which was, however, not significantly different from unity (P = 0.08). In additional analyses, Clade 2 did not have an exceptionally high dN/dS ratio compared to other clades and so has probably not been subject to positive selection.
Expression
In addition to EST and full-length cDNA sequencing projects, the Arabidopsis transcriptome has been characterized by large-scale microarray and MPSS investigations (Brenner et al. 2000; Meyers et al. 2004; Lu et al. 2005). We identified ESTs and full-length cDNAs corresponding to KI and cellular AtULP genes in large public databases. Each of the eight cellular AtULP genes had between 1 and 30 corresponding ESTs, and all but two (At4g33620 and At4g00690) had at least one full-length cDNA (table ST3, Supplementary Material online). In contrast, only eight of the KI genes (9%) had a corresponding EST or full-length cDNA, three of which were full-length cDNAs that overlapped only a small fraction of the gene. Data from a high-density whole-genome tiled oligonucleotide array (Yamada et al. 2003) confirmed this pattern, supporting expression for only 5 of 97 KI genes (5%) compared to 3 of 8 cellular AtULP genes (38%; table ST3, Supplementary Material online).
We also compiled microarray measurements of KI and cellular AtULP gene expression levels in various Arabidopsis tissues using GENEVESTIGATOR, a public Web interface and standardized database of Affymetrix GeneChip data (Zimmermann et al. 2004), which consolidates several large public databases (Craigon et al. 2004; Barrett et al. 2005; Parkinson et al. 2005). Because most KI genes have multiple high-identity copies, many (28 of 61) probesets corresponding to KI genes were ambiguous; however, the ambiguity was mainly restricted to closely related KI genes and both ambiguous and nonambiguous probesets gave similar results (table ST6, Supplementary Material online). Whereas all eight cellular AtULP genes had expression levels significantly greater than background in most tissues (P < 0.06), maximum KI signal intensities were typically lower and not significantly above background in any tissue (Pina et al. 2005). The tissue-wide intensities for KI and AtULP probesets were 103 ± 17 and 597 ± 178 (normalized units; mean of inflorescence, rosette, and roots ± standard error), respectively. The majority (six of eight) of cellular AtULP genes had signal intensities which were 177%–1779% lower in pollen than their tissue-wide means, but two, At4g33620 and At4g15880 (ESD4), had signals that were, respectively, 364% and 141% higher in pollen. Conversely, most KI probesets had signal intensities significantly higher in pollen than their tissue-wide mean (two-tailed paired t-test, P = 9.2 x 10–5; 281 ± 64% increase). Only 23% of KI probesets had decreased expression in pollen. The signal intensities of ambiguous probesets appeared to increase roughly in proportion to increased ambiguity, as would be expected if multiple loci were contributing to the signal. For instance, Clade 1 was represented by five probesets with fourfold redundancy (each probeset hybridizes to the same four KI genes which formed a 97% nucleotide identity cluster) and six probesets with twofold redundancy which, respectively, had average pollen signal intensities of 667 and 202 and average overall signal intensities of 394 and 157. Furthermore, even if only nonambiguous KI probesets were considered, their signal intensities in pollen remained significantly elevated compared to their tissue-wide means (two-tailed paired t-test, P = 2.1 x 10–4). Finally, elevated expression in pollen was supported by lower but still elevated levels of expression in the stamen and inflorescence generally, as recorded by a larger number of microarrays (two for pollen, eight for inflorescence), typically with lower standard error (table ST6, Supplementary Material online).
Although microarrays remain the most widely used technology for performing large-scale analyses of expression patterns, the technology is hybridization based and so, as illustrated by KI, has an inherently limited ability to distinguish weakly expressed genes from the background. MPSS, which involves sequencing millions of short (17–20 bp) cDNA signatures, is both quantitative and sensitive enough to detect transcripts at concentrations as low as three to five transcripts per million (TPM; Brenner et al. 2000). MPSS was recently used to sequence a set of over 36 million 17-bp signatures (268,000 unique signatures) in 14 mRNA libraries from various Arabidopsis tissues, mutants, and treatments (Meyers et al. 2004). We compiled signatures from this set corresponding to KI and mudrA genes in KI-MULEs and to cellular AtULP genes (table ST7, Supplementary Material online). Because of the repetitiveness of KI-MULEs, many KI and mudrA signatures were nonunique; however, as with the microarray probes, ambiguities usually corresponded to KI or mudrA genes in closely related MULEs, and both unique and nonunique signatures yielded similar results. The cellular AtULP genes each had several unique sense signatures (2.4 ± 0.3 per gene) at high maximum abundance across all libraries (26 ± 7.4 TPM) and few or no antisense signatures (0.6 ± 0.3 per gene) at relatively low maximum abundance (8.1 ± 1.8 TPM). Conversely, the combined number of sense signatures for all KI and mudrA genes were only seven (five unique) and six (one unique), respectively, with roughly the same number of antisense signatures, eight (one unique) and three (zero unique), respectively. The maximum abundance for sense and antisense signatures for KI were, respectively, 1.4 ± 0.3 TPM and 3.6 ± 0.7 TPM. Considering only sense signatures (including nonunique signatures), the maximum abundance of KI signatures was significantly lower than that for AtULP genes, consistent with the microarray results (two-tailed heteroscedastic t-test, P = 3.6 x 10–3).
To investigate whether KI may be targeted for RNA-mediated silencing, we examined a database of over 2 million sRNAs (over 75,000 nonredundant) derived from Arabidopsis inflorescence tissues and seedlings that were sequenced by MPSS (Lu et al. 2005) and identified sRNAs corresponding to KI, mudrA, and cellular AtULP genes (table ST8, Supplementary Material online). As with the microarray and mRNA MPSS results, many of the sRNA MPSS signatures for KI and mudrA were nonunique and so could have been generated by any of several closely related genes. However, in this case, the redundancy correlates with biological function because, in vivo, sRNAs presumably target all sequences with which they are able to hybridize. Whereas only two cellular AtULP genes had even a single sRNA, these having low abundance (two and three transcripts per quarter million [TPQ]), KI and mudrA genes had a large total number of sRNA signatures (24 ± 3 and 18 ± 2, respectively) at high abundance (57 ± 6 TPQ and 50 ± 6 TPQ, respectively; excluding the outlier KI-At4g08340 which had an abnormally high abundance of 1512 TPQ due to a nonunique signature which also matches a very highly expressed, unrelated chloroplast gene).
Mobility
We screened for insertions in wild-type Arabidopsis and various mutants (Columbia-0 background) with elevated transposition rates—ddm1, cmt3, met1, and cmt3 met1—using TD, a modification of the amplified fragment length polymorphism technique (Korswagen et al. 1996; Wright et al. 2001). We detected a single new insertion in a cmt3 met1 plant, which we verified by sequencing its termini and the flanking genomic DNA (fig. 3; figure SF4, Supplementary Material online). The terminal sequence uniquely identified the element as the MULE-KI-At2g12100. The flanking genomic sequence contained a perfect 9-bp TSD and mapped to the short arm of chromosome 3, immediately downstream of a geranylgeranyl pyrophosphate synthase gene (figure SF5, Supplementary Material online). RT-PCR experiments showed that the mudrA and KI genes of KI-MULE-At2g12100 have elevated expression in met1 cmt3 compared to wild type (fig. 4).
|
|
|
Discussion |
|---|
|
TOPAbstractIntroductionMaterials and MethodsResultsDiscussionSupplementary MaterialAcknowledgementsReferences |
|---|
Transduplication
We conducted a genome-wide survey of Arabidopsis transduplicates, first applying an accurate procedure we previously developed to identify MULEs and then identifying cellular CDs within these elements. Because many transduplicates do not contain CDs (e.g., approximately two-thirds in rice MULEs; Juretic et al. 2005
), this approach is limited in sensitivity, but it also has key advantages. (1) CDs are compact, highly conserved sequences corresponding to three-dimensional protein-folding units; therefore, sequences similar to CDs are originally derived from genuine coding sequences, not potentially misannotated genes. (2) Since most transposon-related CDs are known, CDs derived from transposon insertions can be filtered out. (3) CD truncations can be identified with high confidence and, along with frameshifts and premature stop codons, are strong indicators of gene disablement (Harrison et al. 2005). (4) Most importantly, in transduplicates that encode functional proteins maintained by selection, it is possible to detect CDs long after their nucleotide sequences and the less highly conserved regions of their amino acid sequences become too divergent from the paralogous cellular genes for any similarity to be detected.
The results of our survey indicate that the general characteristics of MULE-mediated transduplication in Arabidopsis (a eudicot) are similar to those previously documented in rice (a monocot), implying that its mechanism may be widely conserved in higher plants. Eighty-one percent of transduplicated CDs in Arabidopsis (excluding KI; table 1; table ST2, Supplementary Material online) and 83% of expressed transduplicated CDs in rice (Juretic et al. 2005) are truncated by more than 30%, suggesting that they may be pseudogenes (Harrison et al. 2005). Seventy-eight percent of transduplicated CDs in Arabidopsis (excluding KI; table 1; table ST2, Supplementary Material online) and 64% of transduplicates in rice (Juretic et al. 2005) are single copy, and virtually all have fewer than 10 copies in each organism, suggesting that transduplicates do not usually convey a direct selective advantage on the corresponding MULEs. Interestingly, in both Arabidopsis and rice, cellular genes encoding DNA-binding and transcription factors appear to have frequently been the targets of transduplication (table 1; table ST2, Supplementary Material online; Juretic et al. 2005).
KI Sequence Evolution
Despite the many similarities between Arabidopsis and rice transduplication, there is one key difference. KI is a large family of Arabidopsis transduplicates with unique characteristics. Unlike most transduplicates, which have low copy number and contain truncated CDs, there are 97 KI genes with intact peptidase C48 domains as well as at least 30 with truncated domains (tables ST3 and ST4, Supplementary Material online). The KI family is also unusual in its diversity. Most other identified transduplicates have high nucleotide sequence similarity to their putative cellular gene paralogs (over four-fifths have greater than 70% identity), indicating that they formed recently (table 1; table ST2, Supplementary Material online). In contrast, KI has split into nine highly diverged clades with as much evolutionary distance between clades as between KI and cellular AtULP genes, suggesting that extant KI genes may have arisen from an ancient transduplication event. This is supported by the presence of hundreds of KI-like sequences in B. oleracea (despite 0.5x coverage; data not shown), consistent with a KI origin prior to the divergence of Arabidopsis and B. oleracea, 15–20 MYA (Yang et al. 1999; AGI 2000). Curiously, rice, which diverged from Arabidopsis approximately 200 MYA (Yang et al. 1999; AGI 2000), contains at least 161 genes with peptidase C48 domains (Bateman et al. 2004); however, preliminary analysis indicates that they are predominantly associated with other DNA transposons (data not shown).
Two alternative hypotheses might explain the large number of intact peptidase C48 domains in KI-MULEs: either KI-MULEs have coincidentally undergone a large expansion recently enough for the domains to have escaped mutation or KI has evolved under selective constraint. While the aforementioned evidence supporting an ancient origin of the KI gene family provides only circumstantial evidence that the second of these possibilities is correct, nucleotide substitution patterns convincingly demonstrate that KI has been subject to selection. Ratios of nonsynonymous (dN) to synonymous (dS) nucleotide substitutions per site that are much smaller, somewhat smaller, or not significantly different from unity indicate, respectively, that many, some, or none of a set of putative coding sequences have been subject to purifying selection (Li, Gojobori, and Nei 1981). Similarly, values significantly larger than unity indicate positive selection. For the KI gene family, both overall and clade-specific dN/dS values strongly support the claim that KI has evolved under purifying selection. The overall dN/dS for intact KI peptidase C48 sequences (0.24) is within the range typically observed for functional Arabidopsis genes (Zhang, Vision, and Gaut 2002) and comparable to that of cellular AtULP peptidase C48 sequences (0.12). Also, eight of nine KI clades have individual dN/dS values significantly smaller than unity (fig. 2; table ST5, Supplementary Material online). Furthermore, the KI gene family is typical of transposon-related genes in that, in addition to putatively functional members, it contains many apparent pseudogenes which have obvious disablements such as truncations, premature stop codons, and frameshifts (table ST3, Supplementary Material online). Because the codon sequences of these pseudogenes have presumably been drifting neutrally since becoming disabled, we would expect that the strength of selection on the functional fraction of KI genes has been underestimated by these dN/dS calculations. It is important to keep in mind that because KI is located in transposons and therefore presumably subject to different selective constraints than cellular AtULP genes (see below), direct comparisons between the two should be interpreted with caution. Nevertheless, the observed pattern of peptidase C48 domain sequence evolution indicates that KI has maintained a protein-coding function which utilizes the peptidase C48 domain.
The conclusion that KI encodes a peptidase is further supported by the conservation of invariant residues in widely diverged KI sequences. Peptidase C48 contains four highly conserved residues: a putative catalytic triad of histidine, aspartate, and cysteine and a glutamine residue predicted to help form the oxyanion hole (Li and Hochstrasser 1999, 2000). There are some reported exceptions, including human adenovirus type 2 and African swine fever virus, which contain peptidase C48 domains with a glutamate and an asparagine, respectively, at the aspartate site (fig. 1; Li and Hochstrasser 1999). The majority of sequences in three KI clades (Clades 2, 3, and 9) encode all four invariant residues, and the active site sequences appear to be more highly conserved than other parts of the domain, consistent with the preservation of catalytic activity in these gene products (fig. 1; figure SF2, Supplementary Material online). To varying degrees, most other KI clades also appear to have maintained some invariant residues.
Expression
The Arabidopsis transcriptome has been exceptionally well characterized by EST and full-length cDNA sequencing, whole-genome microarrays (Edgar, Domrachev, and Lash 2002; Brazma et al. 2003; Craigon et al. 2004), tiled oligonucleotide arrays (Yamada et al. 2003), and MPSS (Meyers et al. 2004; Lu et al. 2005). We used these resources to compile a detailed picture of the expression pattern and sRNA matches of KI and, for comparison, associated mudrA genes and cellular AtULPs. The repetitiveness of KI-MULEs presents an inherent complication in the interpretation of microarray and mRNA (but not sRNA) MPSS results because most of the microarray probesets and virtually all MPSS signatures match multiple genomic locations. However, in many cases, most or all of the ambiguous locations are within closely related KI (or mudrA) genes. Therefore, although it is impossible to determine which KI (or mudrA) gene among a set of matches contributes to signal amplitudes, we can be reasonably confident that the observed expression patterns are due to some KI (or mudrA) gene in the set. The low level of KI and mudrA expression further complicates the interpretation of the microarray (but not the MPSS) data because many individual signals are not significantly above background.
Despite these complications, our results across all data sets and most loci consistently show that KI expression and sRNA patterns are similar to mudrA and different from cellular AtULP genes. EST, cDNA, tiling array, microarray, and MPSS data all indicate that whereas cellular AtULP genes are generally expressed at significant levels, KI and mudrA genes are expressed at low or undetectable levels (tables ST3, ST6, ST7, and ST8, Supplementary Material online). Microarray data suggest that this pattern is reversed in pollen, where the expression of six of eight cellular AtULP genes is roughly 2- to 17-fold lower than in other tissues while the expression of KI genes is increased on an average of threefold (table ST6, Supplementary Material online). Interestingly, the two cellular genes with increased expression in pollen (ESD4 and At4g33620) represent widely diverged branches of the cellular AtULP phylogeny (figure SF3, Supplementary Material online), which would be consistent with a separation of functional roles like that of S. cerevisiae ULP1 and ULP2. These two genes may play specific specialized roles in pollen.
The MPSS results confirm these trends and provide a higher resolution picture, showing that KI and mudrA genes generate few sense transcripts and a disproportionately large number of antisense transcripts (table ST7, Supplementary Material online). The 97 KI genes with intact peptidase C48 domains have only 14 mRNA signatures in total, roughly the same number of which are antisense and sense. The mean abundance of the antisense transcripts is almost threefold higher than that of the sense transcripts. The opposite pattern is true of the eight cellular AtULP genes, which have a total of 26 mRNA signatures. Only one-quarter of these signatures are antisense, and these have mean abundance more than threefold lower than the sense transcripts. Only one KI gene (At1g40078) has a sense signature with abundance (16 TPM) greater than that of the least abundant primary signature of any cellular AtULP (At4g33620, 14 TPM).
sRNAs, which are generated by the cleavage of double-stranded RNAs including long RNA hybrids (e.g., a sense and an antisense transcript) and short RNA "hairpins," target complementary genomic DNA sequences for transcriptional silencing by heterochromatin formation, and target complementary mRNA sequences for posttranscriptional cleavage (Baulcombe 2005). The sRNA MPSS data show that, while there are virtually no sRNAs for cellular AtULP genes, KI and mudrA match numerous, highly abundant sRNAs (table ST8, Supplementary Material online). Only a single KI gene (At5g33259) has no sRNA, and each KI has an average of approximately 24 sRNAs with abundance 57 TPQ. The large quantity and abundance antisense transcripts and sRNAs are consistent with sRNA-mediated gene silencing of KI and mudrA.
Mobility
MULEs and other DNA transposons have low transposition rates in wild-type Arabidopsis and elevated rates in DNA and histone methylation mutants (Singer, Yordan, and Martienssen 2001; Lippman et al. 2003). There is to our knowledge no documented case of a non-TIR MULE (e.g., KI-MULEs) being mobilized. Several of our results provide circumstantial evidence of recent KI-MULEs transposition. (1) KI is predominantly located in mudrA-containing MULEs (i.e., potentially autonomous MULEs) which must presumably have been recently mobile in order to have maintained these mudrA ORFs (see also the discussion of nonphenotypic selection, below). (2) Many KI-MULEs have perfect target site duplications, indicating that they are recent insertions. (3) Upon replicating, duplicated transposons are identical, so divergence between paralogous transposon sequences is a measure of time since duplication. The nucleotide sequences of several KI genes are nearly identical, for example, two clusters (six sequences in total) have 100% identity and eight clusters (25 sequences from six clades) have 99% identity (figure SF3, Supplementary Material online). While a few of these copies may have arisen through segmental duplication or other cell-mediated mechanisms (e.g., KI-At1k1483 and KI-At1g40078 form one of the 100% identity clusters but appear to have been duplicated in an inverted segmental duplication), differences in TSDs and flanking sequences show that the majority resulted from replicative transposition. This strongly suggests that a diverse group of KI-MULEs were recently mobile; however, it does not exclude the possibility that transposition has since been silenced.
Transposons may be epigenetically silenced by sRNA-directed heterochromatin formation, which involves DNA and histone methylation, primarily via DDM1 (decrease in DNA methylation 1)–dependent histone H3 lysine 9 methylation (H3mK9) as well as MET1 (METHYLTRANSFERASE1)-dependent cytosine methylation at CG sites and, in plants, CMT3 (CHROMOMETHYLASE3)-dependent cytosine methylation at non-CG sites (Bartee, Malagnac, and Bender 2001; Miura et al. 2001; Gendrel et al. 2002; Kato et al. 2003; Lippman et al. 2004). Like most Arabidopsis DNA transposons, KI-MULEs are concentrated in heterochromatin at the pericentromeres and at two isolated islands on chromosomes 4 and 5 (table ST1 and figure SF1, Supplementary Material online; AGI 2000; Lippman et al. 2004). Transposons in these regions, as well as those in cytologically defined euchromatin, have been shown to be subject to H3mK9 and cytosine methylation (Lippman et al. 2003; Zilberman, Cao, and Jacobsen 2003; Lippman et al. 2004). This is consistent with our results, which indicate that both KI and mudrA genes in most KI-MULEs are associated with antisense transcripts and sRNAs and are only weakly expressed.
To test whether KI-MULEs remain transpositionally competent, we screened for insertions in wild-type Arabidopsis and various mutants (Columbia-0 background) known to have elevated transposition rates—ddm1, cmt3, met1, and cmt3 met1—using TD (Korswagen et al. 1996; Wright et al. 2001). Although previous studies found increased transposition of a MULE and CACTA elements in ddm1 mutants (Miura et al. 2001; Kato et al. 2003; Lippman et al. 2003), we did not detect KI-MULE mobility in these mutants (figure SF4, Supplementary Material online). However, we did detect a single new insertion of the MULE containing the KI gene At2g12100 in a cmt3 met1 background. This confirms that at least one KI-MULE remains capable of mobility and provides the first experimental evidence of non-TIR MULE mobility (fig. 3). The mobility of this MULE may be related to the expression levels of its mudrA and KI genes, which are elevated in the met1 cmt3 background compared to wild type (fig. 4).
Interestingly, At2g12100 belongs to a cluster of four closely related KI genes: it has 99%, 97%, and 96% nucleotide sequence identity with the peptidase C48 domains of At1g45090, At2g16180, and At2g05450, respectively. This suggests that the corresponding MULEs were recently mobile in wild type. Although we found no evidence in public databases (i.e., ESTs, full-length cDNAs, tiled oligonucleotide arrays, microarrays, and MPSS mRNAs) that At2g12100 is expressed in wild type, it has no obvious disablement, contains 99.5% of the peptidase C48 domain including all four invariant residues, and the corresponding mudrA gene (At2g12150) contains 100% of two mudrA-related CDs (table ST3, Supplementary Material Online).
Is KI Selfish?
Selfish genetic elements have a unique mode of survival in the genome. Cellular (i.e., nonselfish) elements are selected through their contribution to beneficial phenotypes which increase reproductive success (i.e., phenotypic selection). Selfish elements are also subject to phenotypic selection, but the phenotypes associated with selfish elements are probably deleterious or neutral in most cases because new insertions are likely to jeopardize the function of nearby sequences (Kidwell and Lisch 2001). Thus, phenotypic selection likely works to remove selfish elements rather than conserve them. But selfish elements do not require phenotypic selection to survive, and can even escape mild negative phenotypic selection, because of their ability to self-replicate. By duplicating with sufficient frequency, at least one copy of a selfish element (in an interbreeding population of host organisms) can continually escape disablement and remain able to self-replicate. This process, which has been termed "nonphenotypic" selection, inherently selects for self-replication. (Doolittle and Sapienza 1980; Orgel and Crick 1980; Hurst and Werren 2001; Brookfield 2005).
Unlike other selfish elements, the evolution of eukaryotic DNA transposons is also shaped by the constraint that, to catalyze self-replication, transposase-encoded proteins must presumably act in trans after being imported into the nucleus. Thus, transposons that encode functional transposases (i.e., autonomous elements) may be no more likely to be replicated by their products than related nonautonomous elements. This leads to the replication and accumulation of nonautonomous elements, which often significantly outnumber autonomous elements, and can result in decreased rates of autonomous transposition and eventually the complete silencing of DNA transposon families (Brookfield 2005).
Thus, MULEs evolve under a tension of opposing evolutionary forces. Whereas nonphenotypic selection favors elevated replication rates, new transposon copies can insert into functional cellular sequences causing negative phenotypic selection. As a result, not only do hosts maintain transposition silencing systems such as sRNA-directed transcriptional and posttranscriptional gene silencing, but transposons also evolve self-regulatory mechanisms such as tissue-specific promoters to maximize their reproductive success while minimizing deleterious effects (Kidwell and Lisch 2001). For instance, promoters in autonomous maize MULEs contain nested sets of pollen-specific motifs, and reporter gene expression is increased more than 20-fold in pollen compared to leaves (Walbot and Rudenko 2002). Preferential accumulation in heterochromatic regions with low gene density may also limit the deleterious consequences of transposition (Kidwell and Lisch 2001).
KI has the expected characteristics of a family of selfish genes. Like mudrA and unlike cellular AtULP genes, KI genes are targeted for silencing by numerous abundant sRNAs, are located primarily in heterochromatic regions, are not expressed at high levels, and are preferentially transcribed in pollen. Because KI genes have replicated to high copy number within potentially autonomous MULEs, nonphenotypic selection is sufficient to explain the observed conservation and dN/dS values. Although transposon-related genes are sometimes co-opted to perform cellular functions, these sequences subsequently lose their mobility, which is no longer required for conservation and may be a liability because mobilization could lead to deletion or inactivation of a beneficial cellular function (Cowan et al. 2005). However, KI-MULEs generally have intact termini, contain mudrA genes and so are potentially autonomous, and at least one has retained the ability to transpose. These observations may suggest that KI evolution has been predominantly shaped by selfish selective forces; nevertheless, the possibility remains that a contribution has been made by selection for advantageous phenotypes that KI may contribute to its host.
Function
All known proteins containing the peptidase C48 domain are small ubiquitin-like modifier (SUMO)–specific proteases (Li and Hochstrasser 1999). SUMO is a peptide tag that modulates the function of target proteins in diverse processes, including nucleocytoplasmic transport, signal transduction, cell-cycle progression, stress response, and transcriptional regulation (Novatchkova et al. 2004; Hay 2005). Arabidopsis contains nine SUMO genes, at least one of which is a pseudogene (Novatchkova et al. 2004). Among the four which show significant levels of expression, two appear to be involved in stress response signal transduction pathways (Kurepa et al. 2003; Lois, Lima, and Chua 2003). The SUMOylation of transcription factors usually correlates with transcriptional repression (Gill 2005). SUMO-specific proteases (i.e., Ulps) function as endopeptidases to activate SUMO from its inactive precursor and isopeptidases to deconjugate it from target proteins. Yeast encodes two Ulps which differ in their primary function and localization: Ulp1 is an endopeptidase and localizes to the nuclear periphery and Ulp2 is an isopeptidase and is distributed throughout the nucleus (Li and Hochstrasser 1999, 2000). Interestingly, although complete ULP1 deletion is lethal, nonlethal mutations have been reported that result in the proliferation of yeast plasmids, a type of selfish element (Dobson et al. 2005). Furthermore, SUMO plays a role in plant defense responses, and some viruses and pathogenic bacteria encode Ulps which disrupt host defense mechanisms (Xia 2004).
Consistent with our results, previous reports have noted the relatively large number of ULP-like genes and pseudogenes in Arabidopsis (Kurepa et al. 2003; Murtas et al. 2003; Novatchkova et al. 2004). Although Kurepa et al. (2003) identified only four of the eight cellular AtULP genes found in this study (which they classified as ULP1-like, as well as eight KI genes, which they classified as ULP2-like), Novatchkova et al. (2004) suggest as candidate functional genes exactly the same set of eight cellular AtULP genes as we identified. Our results provide strong evidence that this set of eight genes, which form a separate phylogenetic group from KI, are the only cellular Arabidopsis ULPs (figure SF3, Supplementary Material online). They each contain an intact peptidase C48 domain, are annotated as nonpseudogenic ORFs (TIGR5 annotations), are located in gene-rich euchromatic regions, are not associated with MULE features such as nearby mudrA ORFs or flanking MULE termi